Turn Source Images Into New Art With Nano Banana 2 Image To Image

Last verified: August 2, 2026

A brand designer has a strong campaign image but needs a fresh visual language without losing the idea that made the original work. This workflow turns that moment into a focused creative decision: define what should remain recognizable, describe the new aesthetic, and generate a polished variation.

Nano Banana 2 Image To Image is powered by Gemini 3.1 Flash Image, Google's native visual model for image generation and conversational editing. At the model level, it accepts text and image context; this page channels that multimodal foundation into source-driven transformations with a prompt, uploaded files, resolution, and aspect-ratio selection.

For creators, the game-changing advantage is the ability to art-direct a transformation rather than settle for a loose filter effect. Google frames the model around higher-fidelity editing and stronger adherence to layered instructions, helping a clear aesthetic brief carry through composition, texture, lighting, and finish.

Explore More Image To Image

Capability Snapshot

Verified Run Snapshot

The documented inputs, controls, output, and expected generation cost at a glance.

Workflow

Image-to-image transformation

Required fields

Prompt and Resolution

Supported

Source upload

Multiple source files supported

Resolution

1K, 2K, or 4K

Aspect ratio

1:1, 2:3, 3:2, 3:4, 4:3, 9:16, or 16:9

When Source-Led Transformation Is the Better Fit

Use these decision factors to choose between this workflow and a different creative process.

Starting material A strong choice when one or more existing visuals should guide the new image. Use text-to-image when no source asset exists and the composition can begin from a blank canvas. Restyling photos, artwork, campaign assets, and visual references
Editing precision Suited to prompt-led reinterpretation where the goal is a coherent new variation. Use a manual layer-based editor for exact masks, pixel-level retouching, or deterministic placement. Art direction, concept development, and broad aesthetic changes
Delivery format Fits projects that can use the available 1K, 2K, or 4K settings and one of seven listed ratios. Use a custom-canvas workflow when the deliverable requires an unlisted dimension or print-specific layout. Square, portrait, landscape, vertical, and widescreen digital assets
Iteration strategy Best for deliberate briefs because each run returns one image and is expected to use 15 credits and take 91 seconds. A lighter sketching workflow may fit better when dozens of rough concepts must be explored rapidly. Planned transformations with a defined visual target

Choose this workflow when the source already contains something worth preserving and the creative objective is a thoughtfully art-directed variation rather than an exact manual edit.

Bring Several Sources Into One Brief

Upload multiple source files before submitting the transformation. Use each file for a clear purpose, such as the main subject, supporting objects, or aesthetic direction, rather than adding references that compete for attention.

Set the Deliverable Shape Up Front

Choose 1K, 2K, or 4K resolution and select one of seven available aspect ratios before generation. This makes it possible to frame the result for a square post, portrait creative, or widescreen asset without relying on an undocumented custom-size control.

Strengthen the Brief Before Submission

Use the available AI prompt helper while drafting the required prompt. Review the final wording so it clearly separates the subject elements to preserve from the style, lighting, composition, and atmosphere you want to change.

From Source Files to a Finished Variation

Complete the transformation in four practical steps, from art direction to download.

1

Step 1: Write the Transformation Brief

Describe the intended result, including what should remain recognizable and what should change in the style, setting, composition, color, or lighting. The AI prompt helper is available if you want assistance shaping the required prompt.

2

Step 2: Upload the Source Files

Add the images that should guide the transformation. Multiple source files are supported, so choose a focused set that communicates the subject and aesthetic direction without unnecessary visual conflicts.

3

Step 3: Select Resolution and Framing

Choose 1K, 2K, or 4K resolution, then set the aspect ratio to 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, or 16:9 based on the intended placement.

4

Step 4: Generate and Download

Click "Generate" to create the single high-quality image. Inspect the finished composition, subject details, and any visible text, then download the final output.

Preflight Checks for Nano Banana 2 Edits

Verify these four points before committing 15 credits to an expected 91-second generation.

Before you generate, verify what must remain unchanged

Cause: A broad style request can leave the preservation priorities ambiguous.

Fix: Name the stable elements first, such as facial traits, product shape, pose, logo placement, or architectural structure, then describe the intended transformation.

Retry: Submit after the prompt clearly separates preserved details from changed details.

Before you generate, verify the source files support one direction

Cause: References with conflicting subjects, lighting, or visual styles can make the intended anchor unclear.

Fix: Remove unnecessary files and state the role of each remaining source in the prompt, prioritizing one main subject or composition.

Retry: Retry only after every uploaded file contributes to the same planned result.

Before you generate, verify the aspect ratio matches the composition

Cause: A portrait subject placed in a wide ratio, or a broad scene placed in a narrow ratio, can force an awkward crop.

Fix: Choose the destination ratio first and describe framing that suits it, such as close portrait, centered square, full-body vertical, or panoramic landscape.

Retry: Generate after the ratio and written composition describe the same frame.

Before you generate, verify small text and fine faces can be reviewed

Cause: Google notes that image models can still struggle with small faces, exact spelling, and very fine details.

Fix: Shorten embedded copy, use larger lettering, request a closer crop for important faces, and plan to proofread or inspect the output at full resolution.

Retry: Retry after simplifying the fragile detail or giving it more visual prominence.

Frequently Asked Questions

If I'm a creator, when is Nano Banana 2 Image To Image the right workflow?

Choose it when you already have a photo, illustration, product visual, or other source asset and want to reinterpret it as a new style or variation. A text-only workflow is usually a cleaner fit when there is nothing to preserve from an existing image.

If I'm a creator, what should my transformation prompt include?

State the subject details that must remain recognizable, followed by the desired style, composition, lighting, palette, mood, and finish. If the image needs lettering, provide the exact copy and explain where it should appear.

If I'm a designer, can I upload more than one source image?

Yes. The documented workflow supports multiple source files, but the supplied tool context does not specify a maximum upload count. Use the smallest focused set that communicates the subject and intended aesthetic clearly.

If I'm a designer, can the result include readable words?

Google's model materials highlight more reliable in-image text and localization, but they also advise checking spelling and fine details. Keep copy concise, specify the exact wording, and proofread the generated image before publishing.

If I'm preparing social assets, which aspect ratio should I choose?

Use 1:1 for square placements, 9:16 for vertical stories or mobile-first creative, and 16:9 for widescreen banners or presentations. The 2:3 and 3:4 ratios suit portrait layouts, while 3:2 and 4:3 provide landscape framing.

If I'm planning iterations, can one generation return several variations?

No. The documented output count is one image per generation. Each run is expected to cost 15 credits and take 91 seconds, so refine the source selection, prompt, resolution, and aspect ratio before submitting another variation.

If I'm publishing client work, what usage rights should I check?

The supplied tool context does not establish a commercial-use license or transfer of rights. Confirm that you have permission to use every source file and review the applicable service and model terms before delivering or selling the result.

If I'm checking image provenance, does the model add an AI watermark?

At the model level, Google states that generated images include SynthID. The tool context does not specify whether the downloaded file displays a visible mark or carries additional metadata, so do not promise a particular watermark appearance.

If I'm fixing a weak result, can I use a negative prompt or fixed seed?

No negative-prompt field or seed control is documented for this page. Rewrite the standard prompt in positive, concrete terms by describing the intended subject, framing, style, lighting, and details, then retry after checking the uploaded sources and output settings.

References

Sources and citations used to support the content provided above.

Updated: 2026-08-02 16:07:10 5 Sources

ai.google.dev

Source Link
https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image

blog.google

Source Link
https://blog.google/innovation-and-ai/technology/developers-tools/build-with-nano-banana-2/

deepmind.google

Source Link
https://deepmind.google/models/gemini-image/flash/

blog.google

Source Link
https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/

ai.google.dev

Source Link
https://ai.google.dev/gemini-api/docs/image-generation